ge

the upper level managing particlenet like update steps (like edgeconv and subedge, can mostly use m.edgeconcat to decide if you should concat or replace the output (concat:like particlenet, replace:probably better for autoencoder)

Arguments:

• g: a grap object
• m: constant variable, created for example by getm()
• k=4: the k in topK

Returns

• g: the updated grap value